4.1. Statistical Evaluation of Geochemical Data
Although there is a remarkable difference in concentrations of several studied elements between the deep and shallow samples, considering the whole database, each element exhibits a unimodal distribution (histogram). Most distributions are close to normal, while some of them (Mg, K, Rb, Hf) are highly asymmetric and suggest lognormal distribution. As most applied multivariate statistical methods presume normal distribution, these variables were log-transformed in a common way.
In the first step of the multivariate statistical evaluation, principal component analysis for all studied elements was fulfilled what resulted in five components (eigenvalue > 1) (
Table 2). All these components can be interpreted geochemically as an accumulation process of the corresponding sets of elements. The first component (PC1-1, ~38% of the total variance) has high correlation coefficients with the REEs as well as Ti, Al, and Fe among many other elements. As there are elements of significantly different geochemical behaviours, this set probably suggests the cumulative effect of more REE-accumulation processes.
Interpretation of the second component (PC1-2, ~23% of the total variance) is much more straightforward. Here Cu, Pb, Zn, As, Cd and Fe appear, a set of elements of similar geochemical behaviour. Most of them, but Fe prefer to bond with sulphur and so define the chalcophile elements based on the classical Goldschmidt classification [
8]. In fact, Fe also commonly occurs as sulphide (e.g., pyrite) and so is familiar with other elements of PC1-2. Even if no single sulphide mineral grain was identified in the red muds by any phase analytical method, the common behaviour of these elements is typical even in bauxites [
9]. Nevertheless, the accumulation tendency of these elements is suggested to be totally independent of that typical for the REEs (
Table 2).
The third component (PC1-3, ~9% of the total variance) contains the high field strength elements (HFSE—Zr, Nb, Ta, Hf, Ce), which are known to behave in a similar way in numerous geochemical processes both under igneous and sedimentary conditions. Among the mineral phases of the studied red mud samples zircon, monazite and apatite are the major containers of the HFSE elements. All these refractory phases usually represent the igneous and/or metamorphic source rocks of the original bauxite. Their common behaviour in the red mud suggests that this set of elements could remain immobile not only during lateritic weathering that is the formation of the bauxite but also during the industrial treatment (Bayer process) and under diagenetic conditions inside the pitfalls. The only rare earth element in PC1-3 component, Ce represents the ancient igneous and/or metamorphic source, as well. Because Ce appears not only in a trivalent ionic form, like all other REEs, but also can form a smaller sized, 4+ ion, its compatibility to two different groups of trace elements (and minerals) is reasonable. On the other hand, numerous observations suggest that the major REE minerals in bauxites of diverse origins are zircon, monazite and apatite [
9,
10], which is not the case for the Almásfüzitő red mud. Here, the REEs but Ce have no genetic relationship with these phases as shown by their low correlation coefficients with PC1-3.
The geochemical explanation for PC1-4 (~ 6% of the total variance), which contains Mg, K, and Rb, is more questionable. Nevertheless, as the only potassium phase present in all studied samples is illite, this component probably represents the clay mineral content of the mud. Illite structure commonly contains Mg in the octahedral position, while Rb replaces potassium [
11]. As correlation coefficients of all REEs are very small with PC1-4, the clays probably do not play any role in REE accumulation. Finally, the common behaviour of As, P and W in PC1-5 (~5% of the total variance) conforms with their similar ionic forms (AsO
43−, PO
42−, WO
42−) under highly alkaline pH conditions, which is typical in red muds.
The main result of the first step of the principal component analysis is that there are essential and well identifiable geochemical processes, like an accumulation of the HFSE and the chalcophile elements as well as their minerals, which are independent of REE accumulation. Moreover, there is a well-defined set of elements; all appear in PC1-1 component, which seems moving together with the REEs (
Table 2). That is, why in the second step of the PCA (PC2) only the elements listed in the previous PC1-1 are included. Here three main REE accumulating processes can be recognized (
Table 3). In PC2-1 (~28% of the total variance) Ti, the light and some heavy REEs (La, Ce, Pr, Nd, Sm, Eu, Gd, Tb, and Dy) appear. Comparing this information with the results of the phase analysis, the Ti-phases (anatase, rutile, ilmenite, titanomagnetite) are responsible for this process. Partition coefficients between rutile, ilmenite, titanomagnetite and silicate melts (magma) of any compositions are extremely low as suggested by several experiments [
12,
13]. Consequently, these refractory igneous Ti-minerals probably are not responsible for accumulating the REEs in the red mud studied, what for the only candidate is anatase.
In PC2-2 (~27% of the total variance) Ni, Fe, V, Sc as well as the rest of the heavy REEs (Ho, Er, Tm, Yb, and Lu) are collected. Based on the mineralogical data, the main accumulators for this set of elements should be the Fe-phases, goethite, and hematite. These two minerals and especially goethite commonly have a prominent role in fixing the mobile REE already in laterites and bauxites [
14,
15]. Other authors, like Reinhardt et al. [
16] found that formation of the Fe-oxides is incompatible with the enrichment of REEs. Vind et al. [
17] report that in some bauxites from Greece Sc is mainly hosted in hematite, while there in the red muds goethite accumulates Sc with a concentration of about two times more than hematite.
The variables with high correlation coefficient in the last component (PC2-3, ~26% of the total variance) are Sr, -Ca, Al and a selected set of REEs (Nd, Sm, Eu, Gd, Tb, Dy, Ho, Er, Tm, Yb, and Lu). The signs for Ca and Al suggest that the accumulation tendencies of their host minerals are just the opposite that is when Al increases, Ca decreases and vice versa. Keeping also the phase analytical results in mind, cancrinite, the main Al-phase and calcite, the dominant Ca-phase tend to appear separated in the pitfalls. As all REEs have a positive sign, similar to Al, cancrinite should host these elements, while calcite should not, opposite to the common observations [
18].
Based on the two subsequent PCA calculation steps, numerous element accumulation processes act in the red mud deposit could have been explored, three of which are responsible for the accumulation of the REEs. These latter processes can be linked to the major mineral phases of the red mud, which suggests that accumulation of the REEs is in connection with the behaviour of anatase, goethite as well as cancrinite.
Using the three principal components of the second PCA calculation, the red mud samples were classified using hierarchical cluster analysis. Based on these calculations, two main natural sample groups (G1, G2) can be clearly delineated. To understand the difference that is the weight of the most essential REE accumulating processes between these two sample groups, discriminant function analysis was computed. Its result suggests that the function separates the two groups in the best way is D = −0.7 × PC2-1 + 1.1 × PC2-3. The opposite signs show that while for the G1 samples PC2-3 is high and PC2-1 tends to be low, for G2 samples PC2-1 is the significant accumulation process. Consequently, the separation of the Ti-phases (PC2-1) and cancrinite (PC2-3) is the most responsible process for developing different red mud types, at least, concerning their REE accumulating behaviours. As the coefficient of PC2-2, the variable that summarizes the role of the Fe-phases is negligible in the above function, goethite and hematite should not be responsible for the main separation process. It nevertheless does not say anything about the amount of REE accumulated by the three sets of the container minerals.
4.2. Origin of the Major Mineral Phases
The origin and evolution of the major mineral constituents of the red mud, which hold the remarkable portion of the REEs, are significantly different from each other. The most Ti-minerals, but anatase are stable at high temperature and pressure and so are common phases of igneous and metamorphic rocks. As there are no sedimentary environments in which rutile, ilmenite or titanomagnetite would form, these phases must represent the original source rock of the precursor bauxite. Anatase is the stable low-temperature polymorphous variety of TiO
2 and probably represents the bauxite similar to goethite and hematite, what develop during lateritic weathering. Cancrinite nevertheless is not a common mineral in the most igneous rocks or bauxites and most probably developed during the industrial treatment [
19] or following it, inside the pitfall due to diagenetic processes.
Scanning microscopic observations show that cancrinite together with goethite and hematite appear in aggregates of intensively cemented clusters of very fine grains. This special microtexture is rather strange in bauxite, but is very similar to that develops from kaolinite under hyperalkaline conditions in some well-documented experiments (
Figure 4, [
20]). Hungarian bauxites, similar to most karst bauxites have relatively high Si/Al ratio. As a mineralogical consequence, in addition to the common Al-hydroxide phases (boehmite, gibbsite, etc.) they also contain a significant amount of clay minerals, usually kaolinite [
16]. As the red mud samples studied contain no kaolinite at all, one can assume that it reacted with the elevated Na-content of the mud and formed the cancrinite pseudomorphs after kaolinite due either to the Bayer process or under diagenetic conditions.
In the PC2-3 component (process) of the above calculation, Al and Ca appear with opposite signs (Al, -Ca) suggesting that the main Al- and Ca-phases do not form simultaneously in the pitfalls. So, development of the major cement minerals, cancrinite (Al) and calcite (Ca) seem excluding each other. A possible geochemical explanation for this phenomenon is the local difference in Ca/Al ratio of the mud. At places where Al concentration is below a certain threshold, cancrinite cannot crystallize. A more viable reason is the difference in CO
2 fugacity inside diverse regimes of the pitfalls. Results of experimental petrology suggest that at high fCO
2 calcite and another Na-phase (e.g., nepheline) are stable, but cancrinite [
21]. Although the current sampling strategy in the Almásfüzitő area did not allow localization the zone, where carbonization is the most characteristic process, previous data suggest that the red muds in question contain a significant amount of calcite in the topmost horizon of the pitfalls. Formation of calcite replacing cancrinite in the shallowest zones due to the reaction with atmospheric CO
2 may explain the negative correlation between Ca and Al [
22]. Nevertheless, it also suggests that as soon as cancrinite disappears, the shallow horizon is not capable to fix REEs any more, shown by the negative correlation between Ca and the REEs. Such a reduced concentration for each REE is suggested in the upper segment of all studied wells of the Almásfüzitő area (
Table 1).
4.3. Spatial Variability
When plotting all samples according to their position inside the pitfalls, in each 35 well G1 group samples (high in cancrinite) clearly represent the 0–4 m, while G2 group samples (high in Ti-phases) the 4–8 m deep interval. Such obvious discrimination suggests differentiation of the red mud due to its density resulting mineralogical and so chemical zonation inside the pitfalls. Gravitational differentiation as a governing process had to cause cancrinite, a low density (~2.4 g/cm3) mineral tend to move upwards, while heavy minerals, first of all, the Ti-phases (anatase ~3.9 g/cm3, rutile ~4.2 g/cm3, ilmenite ~4.8 g/cm3, titanomagnetite ~5.2 g/cm3) tend to sink. Of course, in accordance with the tight relationship between the REEs and these mineral phases, gravitational differentiation modifies the spatial distribution of the REEs as well. One can assume that other heavy minerals (e.g., zircon, apatite) behave in a similar way, these phases, however, are of a little amount and, consequently, thought to have a minor role in REE accumulation.
The coarse grain fraction of the red mud studied contains not only single mineral grains but also goethite, hematite, and cancrinite as form well-cemented coarse particles. As the Fe-phases have a much higher density than cancrinite does, presence of these aggregates hinders cancrinite moving to the shallow zone of the pitfalls. Nevertheless, Ca enrichment in the upper zone, as well as the increased Fe concentration in the deep, confirms the influence of gravitational differentiation process (c.f.
Table 1).
All information considered the most essential geological processes that determine the spatial distribution of the major mineral phases and so the REEs in the Almásfüzitő pitfalls are the following (
Figure 5). (1) Because of their high density, the Ti-phases tend to subside and accumulate in the deepest horizons. (2) Cancrinite develops inside the pitfalls as a reaction product of kaolinite under the hyperalkaline condition. (3) Cancrinite forms well-cemented aggregates with goethite and hematite. (4) Because of its low density, cancrinite tends to rise, but as an aggregate-forming phase, it subsides instead. (5) At the topmost zone of the pitfall, cancrinite reacts with atmospheric CO
2 and forms calcite.
4.4. Distribution of REEs
In order to understand the accumulation of the REEs in the Ti-phases, goethite, and cancrinite, fractions of a single red mud sample (representing the 4–8 m deep interval) were produced. Four mineral concentrates were studied; the magnetic and non-magnetic fractions of the coarse (>20 microns) grains, as well as a cancrinite-rich and a goethite-rich fraction. According to the semi-quantitative XRD estimation, the proportion of cancrinite decreases from 40–50% down to 20–30%, while that of goethite increases from 10–20% up to 40–50% in the last two fractions, respectively. The rest of both fractions is hematite.
Comparing the REE content of the four fractions, the cancrinite-rich sample is the highest in all REEs followed by the goethite-rich one, while the two coarse grain fractions are the lowest in all REEs (
Table 4). Concentrations in the cancrinite-rich concentrate are similar to or are slightly higher than those of the total sample proposing that cancrinite is the main REE container phase of the red mud (
Figure 6). The non-magnetic coarse fraction is higher in each REE than the magnetic mineral separate. Considering the tight correlation between Ti and the light REEs, anatase, as the only non-magnetic Ti-phase may be responsible for this kind of accumulation. Despite this behaviour of anatase and keeping their low modal proportion in mind, the coarse grains are not important REE container minerals of the red mud studied.
Element enrichment between the major phases can be characterized by enrichment factor for each element so that:
where A and B are the mineral phases studied and X stands for concentration of the element in question. To calculate concentrations in the pure end members based on the mineral mixtures, a linear mixing model was used (
Figure 7) based on the following equation system:
where I = 1, 2 for the two fine-grained mixtures, and a
1 and a
2 are the proportions of cancrinite in the mineral mixtures what are only roughly known from XRD.
From this on, there are two equations with four unknowns (a
1, a
2, X
A, X
B) for each element. Moreover, an equation with an identical form exists for the whole sample so that:
where a
tot is the proportions of cancrinite in the whole sample. The parameter triplet (a
1, a
2, a
tot) can be optimized so that be minimal. For optimization, the trial-and-error method was followed (
Figure 8). Finally, based on the estimated a
1 and a
2, both X
A and X
B can be calculated:
The result of the model was accepted at the lowest available ε (= 34.47), where the Pearson correlation coefficient between the measured and estimated total concentrations is better than 0.999 and the slope of the regression line between the measured and estimated concentration is as good as 1.01 (
Figure 8d.,
Table 5).
The best fit coincides with a1 = 35%, a2 = 62%, and atot = 60%, that is the goethite-rich separate contains ~35%, the cancrinite-rich one contains ~62%, while the original sample contains ~60% of cancrinite. These numbers nevertheless refer exclusively to a two-phase (cancrinite, goethite) system and the unknown amounts of all other minerals decrease the above proportions in the case of the whole mud. The low REE concentration of these diluent phases causes that most concentrations measured in the original mud sample and there in the cancrinite-rich separate are rather close to each other.
The main result of the modelling is nevertheless the list of REE concentrations characteristic for the pure end members, cancrinite and goethite (
Table 6) as well as their ratios, the enrichment factors for each REE. The data give a numerical proof for a previous statement that cancrinite accumulates significantly more of each REE than goethite does. These concentrations are also much higher than those typical are in the original red mud (
Table 6). The enrichment factors can be calculated in an identical way between cancrinite and the coarse grain separate as well. In this calculation, magnetic and non-magnetic fractions were handled together (
Table 6). The values suggest that these mineral phases are of a minor role in REE accumulation.
Evaluating the mode in which cancrinite and goethite accumulate REEs is out of the scope of the present paper. However, it has been known from experimental mineralogical results that the crystal structure of both phases is appropriate to physically adsorb the REEs. The adsorption capacity of goethite is usually very good, but it depends on numerous parameters, like Al-content, crystallinity, temperature, and pH [
23]. Some experiments confirm that goethite adsorbs La very well at a pH > 5 [
24], while [
25] found Sc to adsorb on goethite surfaces, by analyzing its behaviour during leaching experiments. Cancrinite has channels defined by a 12-membered silicate ring in its structure [
26,
27] and has been proposed as the possible hosts of the REEs [
28,
29]. Although it was found hard to measure analytically, Vind et al. [
17] report cancrinite a probable host for Sc in red muds from Greece.
Nevertheless, physical and chemical circumstances are not known in detail yet, as there is not enough experimental result available concerning REE adsorption on the two key minerals of the Almásfüzitő red mud. If physical adsorption has a significant role, adsorption-desorption mechanisms of REEs in the red mud, pure goethite and cancrinite should be studied first in order to find an effective industrial tool to remobilize these elements.